A New Nvidia Patent Speeds Up Game Lighting by Focusing on the Brightest Spots
Nvidia's filing turns a picture into a bumpy 3D surface where the brightest pixels are the biggest targets, then shoots virtual light at it. The sun gets hit most often, which is exactly where a game's lighting needs to look.

- Publication number
- US 2026/0368993 A1(US20260368993A1)
- Applicant
- Nvidia Corporation
- Filed · Published
- Nov 17, 2025 · Nov 19, 2026
- Inventors
- Nathan Morrical
- Status when we published
- Waiting for an examiner
- Document
- 1 claims
- US classification
- 345/426
- Parent application
- Continuation of 18/490,171 (filed Oct 19, 2023)(18490171)
What Nvidia's brightness-hill lighting trick does
Every time a game draws a new frame, which can happen 60 or 90 times a second, your graphics chip has to work out how the scene is lit. A sunny sky is the classic headache. The sun supplies most of the light but covers only a few pixels, so checking pixels at random wastes effort on dark ground and dull sky.
Nvidia's patent turns the picture into a bumpy 3D surface, with each pixel's height set by how much it matters for lighting. Then it fires virtual rays of light at that surface. Tall, steep spots get hit far more often, so the sun gets checked constantly and the shadows barely at all.
In the filing's own example, the sun pixel is about six times more likely to be hit than any other pixel. The result is less grainy lighting for the same amount of work.
… generate, based on the importance values, a three-dimensional (3D) geometric mesh of the image corresponding to interaction probabilities for the one or more pixels, respective heights of cells forming the 3D geometric mesh corresponding to respective interaction probabilities …
Translation: It builds a 3D surface where the height of each cell represents how likely light is to interact with that pixel.
How rays pick pixels off a probability mesh
Lighting a scene realistically means estimating how much light reaches each surface. The usual shortcut is importance sampling (checking the most influential pixels more often than the dull ones). The classic way to do it, inverse transform sampling, turns each row and column of pixel brightness into a running-total curve (a cumulative distribution function) and then uses a binary search (repeatedly halving a list to find a match). The filing says that search behaves unpredictably and is costly at 60 to 90 frames a second.
Claim 1 swaps the search for geometry:
- Score each pixel of an image for importance.
- Build a 3D mesh where each cell's height matches that pixel's probability of being picked.
- Fire simulated light rays at the mesh.
- Read off which pixel locations the rays hit.
In the description, the running-total curves become ribbons of triangles, so steeper stretches of a curve offer more surface for a ray to hit. The slope where a ray lands also gives the probability needed to keep the final lighting math fair. Because the mesh is made of triangles, dedicated ray tracing hardware (the filing names Nvidia's RT cores) can search it quickly.
The filing also describes optional tweaks: keeping only the most influential control points to simplify the mesh, rescaling it to a 0 to 1 range (45 to 50% faster traversal in its tests), and merging similar neighboring rows (20 to 30% faster).
A geometric mesh can be generated based on a set of cumulative distribution functions (CDFs) for various rows and columns of pixels in a texture, and individual rays can be traced against this mesh, with those rays having a higher probability of intersection at a point with greater importance, such as a higher brightness value.
Translation: Virtual light rays are bounced against this mesh so they naturally hit the brightest and most important spots first.
Why this matters for real-time game lighting
For anyone who plays games on a PC, console or streaming service, the payoff would be lighting that looks cleaner at the same frame rate, or the same look at a higher one. Grainy, flickering light comes from a renderer taking too few useful samples, and this approach makes each sample count for more.
The filing says its methods are light enough to run on a client device such as a gaming console, and the same math can also run on cloud servers that stream the finished picture to you. It also describes the idea as usable beyond games, for other image work. The reported gain is up to 70% faster rendering in one example, with noise reduction similar to the traditional method.
This is the 46th Nvidia filing we've tracked in the GPU rendering race since July, adding to work like shadows and reflections chip and memory shortcut for AI images.
Source: US 2026/0368993 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.
Get our take in your Top Stories
Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Nvidia patent news breaks.
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings · 18 sheets
18 drawing sheets from US 2026/0368993 A1 · click any drawing to enlarge
Show the other 10 sheets

No comments yet